2.1 Background

In 2016, Venlo, a municipality in the southeast of the Netherlands, inaugurated a newly constructed office building for use by the municipality. The new municipal building was built following “green” and sustainable principles: In addition to glass and concrete, the north wall of the building is covered with vegetation, and includes a green wall of 2,000 m². Green or living walls allow plants to grow from the vertical structure. The installation of green walls has been associated with an improvement in outdoor and indoor air quality, transforming carbon dioxide (CO₂) into oxygen, and filtering fine particles from outdoor sources of pollution [18]. In addition, the plants serve as natural insulation against heat, cold, and sound [19]. The building is also equipped with state-of-the-art natural ventilation technology. The air enters the building at the top, where it is oxygenated by plants and brought to the bottom of the building, from where the purified air then circulates naturally throughout the building using physical principles rather than mechanical ventilation systems.

While the new building was designed to the highest standards of green building principles, it has not applied for a green building certificate such as BREEAM or LEED. In the very first stages of the building’s design process, the aim was to opt for BREEAM certification. However, at that time, BREEAM did not incorporate building health aspects in its certification process and was solely focused on energy efficiency, and, for example, recycling of building materials. This would have steered the design into an unwanted direction: not enough ventilation (to improve energy efficiency) and recycled, but potentially unhealthy building materials. On top of that, the certification process was deemed expensive, and the municipality thought that money would be better spent on a high-quality interior climate maintenance system, which was indeed installed. The new building has been certified under the “Cradle to Cradle” principle, an attestation to the re-usability of products in the building. In addition, it has won a range of accolades for its sustainable design and performance [20].

In the summer of 2016, 70% of the 1,461 workers of the municipality were moved to the newly constructed office building, within the same city. A round of interviews with high-level officials from the municipality confirms that the selection of “movers” was quasi-random, through selection of teams (or: functional groups), rather than individuals. In other words, relocation decisions were independent from the performance, health, or well-being of individual employees. Three well-defined functional groups did not move. First, all city employees who worked in social services and welfare provision were already located in a separate building before 2016 and stayed there. Second, the complete department of public works, dealing with parks, playgrounds, green maintenance, roads, traffic lights, etc. was housed at a separate location before 2016 and stayed there. Third, in 2015, one year before the new City Hall became operational, the city of Venlo completed the renovation of an adjacent building, previously an industrial site. This building does not have a public entrance or reception desk, which is why the municipality decided to locate the Finance and IT departments in the building, since these departments are mostly internally oriented.

In addition to significant changes in indoor conditions, the organizational office layout also changed. Office space in the previous workplaces was organized in enclosed private offices, shared by several people. In the new building, the office space follows an open office layout. Open offices tend to generate noise complaints among occupants, who can be distracted by high levels of noise and loss of privacy. Indeed, a recent study shows workplace distraction significantly reduces worker productivity [21]. We therefore explore the changes in noise satisfaction in our analysis.

The newly constructed building is located very close to the existing buildings—two buildings are within 150 meters (a tenth of a mile), and equally close to public transit and highways ramps. The third building is located at a distance of 1km (0.6 mile), and next to a local train station, the main commuting mode for employees that use public transit (the new building is close to an intercity train station). Overall, employees that moved to the new building did not experience meaningful differences in commuting patterns, and if anything, conditions for access to public transit get slightly worse.

2.2 Survey design

We received permission from the municipality to send surveys to all of its employees, asking them to complete the survey via email (See S1 Appendix for the text of the invitation sent to the employees). The survey included anonymized individual identifiers, allowing us to build a longitudinal dataset that tracks the responses from the same employee over multiple survey waves. Our sample includes the survey responses of the treatment group, individuals relocated to the new “green” building, and those of the control group, comprising those employees who were not relocated to the new building.

The surveys span the period before and after the relocation. We first sent a questionnaire to all employees one month before the relocation took place, serving as the baseline survey in the analysis. After the relocation, we surveyed all employees three times, six months to two years after the relocation. We surveyed those individuals that were relocated as well as those who remained in their original workplace during the entire sample period. The three surveys cover both the cold and warm seasons (see Fig 1 for the exact timeline of the surveys).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Timing of survey waves.

The figure illustrates the exact timing of each survey wave. The first survey (k = 0) took place before the moving date (August 2016). The second (k = 1), third (k = 2) and fourth (k = 3) took place after the relocated group moved to the new building, in respectively January 2017, July 2017 and October 2018.

https://doi.org/10.1371/journal.pone.0236029.g001

The survey included the module developed by the Center for the Built Environment (CBE) at the University of California, Berkeley, to monitor the perceived environmental conditions of occupants in their workplaces [22]. Since the early 2000s, the CBE survey has been extensively used to evaluate the performance of buildings globally. The core questions in the survey assess occupant (dis)satisfaction and comfort with indoor environmental quality (IEQ) issues, including indoor air quality, thermal comfort, lighting, and acoustics. We asked participants to rate their satisfaction with different aspects of the indoor environment on a 7-point scale, ranging from “very satisfied” to “very dissatisfied,” with a neutral midpoint. In a second set of questions, we asked participants to rate each IEQ dimension on 7-point scales ranging from “support” to “interfere” with their ability to get work done (see Panel A of Table A.1 in S1 Appendix).

The survey included two questions about the health status of individuals. First, we examined changes in the health status based on the prevalence of symptoms related to the Sick Building Syndrome (SBS). This concept is widely examined in the building science and public health literature and refers to “a collection of non-specific symptoms including eye, nose and throat irritation, mental fatigue, headaches, nausea, dizziness and skin irritations, which seem to be linked with occupancy of certain workplaces” [23]. The survey included a question asking whether the subject suffered from SBS symptoms (“Do you regularly have symptoms (e.g. tiredness, headache, eye irritation, nasal congestion, dry throat, dry skin) that disappear when you leave the building?”). In addition, we collected self-reported sick leave data based on the number of days missed due to health reasons in the year before the survey (“How many days were you unable to work this year due to illness?”) in a categorical question where participants were asked to choose between the following options to report the number of sick days: (1) I did not report sick this year, (2) 1 day, (2) 2-5 days, (3) 5-10 days, and (4) more than 10 days.

We measured the job satisfaction of employees based on a series of Likert scales, where respondents rated the frequency with which they experienced job-related emotions (see Table A.2 in S1 Appendix for the complete list of questions). The frequency of the scales ranged from “Daily” to “Never,” and includes the options “A few times a week,” “Once a week,” “Few times a month,” “Once a month,” and “Few times a year or less.” To help measure job satisfaction, we construct an index equal to the first principal component of all survey questions listed in the Table A.2 in S1 Appendix, together with the loadings of the first principal component of job satisfaction.

In addition, the survey included a series of questions that asked participants to grade their (dis)satisfaction with the layout, and furniture and equipment in their workplace, based on the same scales that were used to grade the perceived environmental conditions in the workplace (see Panel B of Table A.2 in S1 Appendix). We created dummies from the original scales that take the value of 1 if the original scale is “Daily,” “A few times a week,” or “Once a week,” and zero otherwise. Finally, the survey included questions about basic demographic characteristics of respondents (i.e., age and gender) and some details of the employee’s employment contract (i.e., working hours, and years at the current organization).

2.3 Descriptive statistics

The response rate of the surveys ranged between 35% and 40%. In the first wave, we gathered 573 valid answers, 585 in the second wave, 569 in the third wave, and 530 in the fourth wave. The median completion time of the survey was 11 minutes. We observe no differences in response time between relocated (treated) and non-relocated (control) employees, suggesting no differences in attention or effort between the two groups (but, of course, our goal is not to empirically assess such differences).

Table 1 shows the demographic characteristics of the relocated and non-relocated workers in wave 1 (i.e. the survey administered before the move). The non-relocated employees are younger, on average, than those in the relocated group, as reflected by the higher percentage of individuals below 31 years old (19% vs. 10%). The gender ratio does not differ between two employee groups.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive statistics sample in first survey wave (Before the move, July 2016).

https://doi.org/10.1371/journal.pone.0236029.t001

Looking at the current contract characteristics of the two groups of employees, we find that the non-relocated individuals are less experienced than those in the relocated group (as reflected in the age difference between the groups). The total working hours do not differ significantly between the treatment and control group. Overall, the pre-trends reported in Table 1 indicate the absence of meaningful differences in the main outcomes of interest—the reporting of SBS symptoms, job satisfaction, and the incidence of sick leave.

Fig 2 provides simple non-parametric comparisons of the metrics that relate to occupant (dis)satisfaction and issues with indoor environmental quality (IEQ). Each chart shows the percentage of dissatisfied respondents on 10 dimensions of IEQ, over the four survey waves. A lower percentage implies a higher satisfaction with that particular IEQ dimension.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Non-parametric comparison of IEQ dimensions.

Summer 2016 survey took place before the move of relocated workers to the new building. Percentage based on the number of respondents that rate the environmental dimensions as larger than four on a seven-point Likert scale—ranging from satisfied (1) to dissatisfied (7).

https://doi.org/10.1371/journal.pone.0236029.g002

Panel A shows the satisfaction metrics for the non-treated sample, i.e. those survey respondents that remained in the same building. The scores on almost all IEQ dimension are exactly equal before and after the moving date, which is in line with expectations—after all, nothing changed for these survey respondents. There are some small improvements in satisfaction with two air quality dimension, but those improvements are not consistent across survey waves, perhaps indicating a seasonal effect.

Panel B shows the IEQ satisfaction for the treated sample. Quite clearly, the satisfaction of relocated employees with IEQ dimensions related to indoor air quality, temperature, and to a lesser extent views and light, increased after the move took place. Concomitantly, employee satisfaction with sound and privacy-related IEQ dimensions decreased slightly.

3.1 Empirical strategy

We use difference-in-difference (DiD) models to estimate the impact of the improvement in building conditions on employees’ perceived working conditions and health status. The DiD research design relies on the assumption that the characteristics of workers who were relocated to the new building changed over time in a way that is comparable to those who were not relocated. To alleviate concerns of potential biases in our results, we estimate our parameter of interest in a regression model with a rich set of fixed effects and time-varying control variables: (1) where Y_it includes the set of outcome variables describing the perceived working conditions and health status of individual i at time t. We include the scales describing the perceived noise, temperature, light, and air quality in the workplace. Finally, we consider a dummy variable indicating whether the individual suffers from SBS.

Our prime parameter of interest is δ, describing the average change in the outcomes (Y_it) after the move for the employees who relocated to the new building. The individual fixed effects (μ_i) should reduce bias resulting from differences between the movers and non-movers. In addition, we include time dummy variables τ_t for each survey wave, non-parametrically adjusting for possible shocks in the city or employer that coincide with the move (e.g., pollution reduction in the city). Finally, we include a set of individual time-varying controls, X_it. The set of controls includes the average working hours per week and the reported scales rating the Office Layout (See Panel A of Table A.1 in S1 Appendix for the full list of scales in this category). ϵ_it is the error term, which might be correlated within individuals. Therefore, we cluster standard errors at the individual level.

We then use an event-study analysis to capture dynamic effects of the new building on the employees. Eq 2 estimates the effects of the relocation separately by year: (2)

Here, the coefficient δ_k describes the effect of working in the newly constructed office k periods after the moving date. Thus, is an indicator for being k time periods relative to the moving date. The reference category is k = 0; hence, the post-treatment effects are relative to the year immediately before the treated individuals were relocated to the new building.

In a final step, we estimate to what extent the changes in each of the environmental scales with respect to their baseline level translate into changes in health status with respect to the baseline: (3) where Health_it takes the value of 1 if individual i reports SBS symptoms at time t, and zero otherwise. Health_ib takes the value of 1 if individual i reports SBS symptoms in the baseline survey. Health_it − Health_ib describes the difference between individual i’s probability of stating SBS status at time t and in the baseline survey b. Similarly, IEQ_its − IEQ_ibs describes the changes in the values reported in the environmental scale s for individual i at time t, with respect to his answers in the baseline survey b. The coefficients of interest, Θ_s, describe how changes in environmental scale s translate into changes in the probability of reporting SBS symptoms. In addition, we include the changes in a set of control variables for building quality X_it − X_ib. Error terms are clustered again at the individual level.

3.2 Ethics statement

The Gemeentesecretaris of Venlo vetted the concept survey designed by Maastricht University in the Summer of 2016. Under Dutch law, this is the highest ranking civil servant of a city, with overall responsibility for all matters pertaining to the city bureaucracy, including privacy issues of personnel. He gave his consent, conditional upon agreement by the city’s privacy controllers and his fellow directors. Consequently, the survey, recruitment scripts, and informed consent protocols were sent to the privacy controllers of the city of Venlo for further revisions and vetting. The privacy controllers gave their consent, and the survey went for a final round of vetting to the Board of Directors of the city bureaucracy. The survey was distributed after the directors gave final permission. The distribution of the survey, and all the communication with the subjects in the study was made directly by members of the local government, via a corporate email account. The survey responses were directly loaded in an online platform, where researchers could access the data for further analysis. No human subject was directly contacted by the research team. The research team had no access to information through which survey participants could be identified.

4.1 Difference-in-differences

In this section, we report the estimated coefficients and standard errors of the coefficients associated with the DiD variable in Eq 1. Table 2 provides the estimation results. Column (1) shows the estimated DiD coefficients, including time dummies, individual-fixed effects, and time-varying controls. We include changes in working hours, perceived quality of furniture, and office layout as time-varying controls.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Estimation results from difference-in-difference model.

https://doi.org/10.1371/journal.pone.0236029.t002

The estimation results indicate that the relocation of employees is associated with a significant decrease in the level of dissatisfaction of perceived environmental quality in all measures, except for noise and privacy. The highest impact associated with the relocation is on the air quality dimension, where the dissatisfaction scale drops 1.4 points on a 7-point scale. In relative terms, when compared to the average value of these scales in the baseline survey for the relocated group, the relocation to the new building improve employee satisfaction with air quality by 32% (1.44/4.50). Similarly, the relocated employees attach a 26% (1.09/4.14) lower value to the scale evaluating whether air quality hinders work.

The relocation of workers to the sustainable building also generates significant improvements in the perception of light quality and temperature. The absolute and relative improvement in the scales of these two parameters is smaller than the changes observed for the air quality dimension. Temperature dissatisfaction among relocated workers drops, on average, by 0.59%, that is 17% compared to the average value of the relocated group before the move. Similarly, the relocation reduces dissatisfaction with the light quality in the building by 0.5%, that is 28% compared to the average value of the relocated group before the move.

Anecdotally, we can confirm the enhanced quality of the indoor environmental conditions in the new building through the use of indoor air quality data from several sensing campaigns organized by the municipality in all of their buildings. While a formal analysis is not possible, given the limited scope of the sensing data, Fig A.1 in S1 Appendix provides a summary of CO₂ levels and maximum temperature in one of the existing buildings (2016) versus the new building (2017), during periods of comparable outdoor conditions. The CO₂ measurements, commonly used in the building science literature to evaluate the performance of ventilation systems, corroborate that the air quality of the “green” building relative to the old building is significantly better. Furthermore, the temperature data indicate that, during hot days (peak day above 30°C), there are substantial improvements in the thermal conditions in the green building, which manages to keep the indoor temperature in an optimal range of 22-24°C (vs. 26-29°C in the old building).

Importantly, we observe significant improvement in the health of individuals. Column (1) of Table 2 shows a decrease of 21.6% in the prevalence of symptoms related to the SBS among the relocated workers (after the move). The relocation of workers generates a substantial drop in the prevalence of SBS symptoms when compared to the baseline probability of reporting SBS symptoms among the relocated employees, by 42%.

4.2 Dynamic effects

We test for the existence of a possible rebound in the improvement of perceived health and well-being experienced by the relocated employees. Evidence from psychology and behavioral economics shows that individuals tend to adapt, in the medium term, to changes in their physical or material conditions [e.g., 24]. Thus, the estimated changes in the subjective assessments presented in the previous section might be biased by an overreaction of individuals in the short term. In addition, the potential material depreciation of the new building might also distort the results.

Fig 3 shows the coefficients () describing the changes in responses across survey waves (k) with respect to the baseline survey for the relocated (blue) and non-relocated employees (gray), together with their associated confidence intervals. Our surveys are taken from six months to two years after the relocation. In addition, the three surveys cover both the cold and warm seasons. Overall, we observe stability in the coefficients describing the changes in health status over time. We find no evidence of a rebound in the estimated changes in the health status of the employees. The estimation results indicate the initial drop in the prevalence of SBS symptoms remains at that level (δ_k=1 = 0.21).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Trends in sick building syndrome.

The figure shows the estimated coefficient of the time dummies denoting the survey waves before (= 0) and after the moving date. The dots represent the point estimates and the bars the 95% confidence intervals. The vertical, dashed gray line indicates the moving date. The set of control variables includes the average hours worked per week and the layout scales.

https://doi.org/10.1371/journal.pone.0236029.g003

4.3 Heterogeneous effects

In this section, we study whether some subgroups are more sensitive to indoor environmental conditions than others. First, we explore gender differences. Current regulations regarding indoor climate conditions in office buildings tend to be based on a thermal comfort model developed in the 1960s. That model optimizes the environmental conditions to satisfy an average male. A recent study in biophysics indicates the existing model significantly miscalculates the metabolic rate of female thermal demand [25]. This is line with many field studies showing that females express more dissatisfaction than males with low temperatures [for a review of the literature, see 26]. In addition, the presence of pre-existing diseases in the respiratory or cardiovascular systems among older population groups might exacerbate the health impacts of certain hazards in the indoor environment (e.g., indoor pollutants) [27]. For the analysis of the differences across demographic groups, we therefore stratify our sample by gender and age.

Table 2, Columns 2 to 6 presents the results of the heterogeneity analysis. Columns 2 and 3 display the results for the two gender subsamples, and columns 4 to 6 show the estimates for the three age groups in our sample. We observe no significant discrepancies in responses to scales in the noise, air, and light quality dimensions across gender or age groups. However, the results for thermal dissatisfaction indicate the drops in dissatisfaction rates associated with the new building are present only among male employees. Relocated women did not significantly adjust their ratings after being transferred to the new building. Similarly, we observe significant changes in thermal dissatisfaction among the older employees only (beyond 30 years old).

When focusing on the health measures, we find the relocation of workers to the new building generates similar drops in the prevalence of SBS symptoms for female and male employees. The estimates suggest the impact of the relocation becomes more significant with the age of employees. We observe significant drops in the probability of reporting SBS among the oldest group of employees only (over 50 years old). The coefficient associated with the group of workers between 30 and 50 years old is marginally significant (i.e., at the 10% level).

4.4 Health and productivity

The relocation to a new building involves significant changes in a variety of factors regarding the workplace of employees. We therefore analyze how the changes in different dimensions of indoor environmental conditions contributed to the drop in the prevalence of SBS symptoms, with respect to the initial situation just before the moving date.

Fig 4 presents the estimated coefficients Θ_s of Eq 3, describing the association between the probability of reporting SBS symptoms and each of the four factors related to the indoor environmental conditions in the workplace (i.e., elements of Θ). The estimation results indicate that poor air quality is, on average, the only significant contributor to the prevalence of SBS symptoms. The presence of perceived deficient air quality is associated with an increase of about 10 percentage points in the odds of reporting SBS in our sample.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Effect explaining sick building syndrome by IEQ dimension.

The figure shows the point estimates of and confidence intervals associated with each ot the elements of vector Θ in Eq 3. The dots represent the point estimates and the bars the 95% confidence intervals. All regressions include time varying controls (contract type), individual and survey-wave fixed effects. * p<0.1, ** p<0.05, *** p<0.01.

https://doi.org/10.1371/journal.pone.0236029.g004

We also include a series of furniture quality factors as regressors in the empirical model, as placebos to construct a falsification test. The absence of significant coefficients associated with these factors supports the hypothesis that the health improvements displayed in this study are mainly driven by an improvement in environmental conditions in the workplace and not by a general building quality improvement.

We then implement a mediation analysis [15; 28] to test for the impact of the relocation on job satisfaction and employee sick leave, where the SBS symptoms act as mediator. For the analysis, we construct a dummy variable that takes the value of one if the individual reports sick, and zero otherwise—focusing on the extensive margin. For job satisfaction, we construct an index equal to the first principal component of all job satisfaction survey questions listed in Table A.2 in S1 Appendix. The index ranges from -3.20 to 14.75.

Fig 5 Panel A shows that effect of the relocation on job satisfaction is mediated by the presence of SBS symptoms. The first, direct path indicates that there is no evidence of a direct effect of the relocation on job satisfaction of employees. However, the estimates from the second, indirect path indicate the presence of a significant average causal mediation effect of 0.11, with a 95% confidence interval that does not include zero [.20, .03]. The increase in job satisfaction corresponds to 4% of a standard deviation in the constructed index.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Impact of relocation on productivity.

The figure shows sick building syndrome (SBS) as a mediator of the effect of the relocation to the new healthy building. The regressions include a dummy variable indicating whether the individual was in the relocated group, and survey-wave fixed effects. * p<0.1, ** p<0.05, *** p<0.01. The mediation equations are: Y_it = τ₁ After Move_t + μ₁ Relocated_i + b₁ SBS_it + b₂ Relocated * After Move_it + ϵ_it1 (4) SBS_it = τ₂ After Move_t + μ₂ Relocated_i + a₁ Relocated * After Move_it + ϵ_it2 (5) Where Y_it denotes the sick building syndrome (SBS) or job satisfaction of individual i in survey wave t.

https://doi.org/10.1371/journal.pone.0236029.g005

Fig 5 Panel B shows that effect of the relocation on sick leave is also mediated by the presence of SBS symptoms. The first, direct path again indicates that there is no evidence of a direct effect of the relocation on the incidence of sick leave of employees. However, the estimates from the second, indirect path indicate the presence of a significant average causal mediation effect of -0.02, with a 95% confidence interval that does not include zero [-.04, -.01]. This result implies that the mediating hypothesis is supported. The relocation of individuals to the healthy building leads to a drop in the prevalence of sick leave of 2%, with the presence of symptoms related to SBS as the main mechanism for explaining the treatment effects.